독일 저작권법상 텍스트⋅데이터 마이닝(TDM) 해석 기준과 시사점 – OLG Hamburg, Urteil vom 10.12.2025 – 5 U 104/24 (Kneschke v. LAION) 판결을 중심으로 –

The rapid proliferation of generative artificial intelligence (Generative AI) is fundamentally reshaping the order of content creation, distribution, and use through learning architectures based on large-scale data collection and automated analysis. Generative AI systems are trained by collecting and analyzing a wide range of copyrighted materials—such as texts, images, and audio—publicly available on the web through crawling and scraping techniques. This process inevitably entails acts of reproduction and use without the prior authorization of rights holders. As a result, the permissible scope and regulatory coherence of copyright law with respect to text and data mining (TDM) have emerged as critical issues across jurisdictions. Against this backdrop, on December 10, 2025, the Hamburg Higher Regional Court (Oberlandesgericht Hamburg) issued a landmark decision in Kneschke v. LAION, explicitly recognizing that the collection and analysis of images conducted in the course of constructing AI training datasets fall within the scope of the TDM provisions under German copyright law. While the court proceeded on the premise that the downloading and analysis of images during dataset construction in principle constitute acts of reproduction under copyright law, it clarified that the unlawfulness of such acts or the establishment of infringement must be assessed step by step through the applicability of the TDM exception and the three-step test. Compared to the first-instance judgment rendered on September 27, 2024, the appellate court’s approach is significant in that it more systematically distinguished and examined, in sequence, the issues of reproduction, the qualification of the conduct as TDM, the validity and effect of opt-out mechanisms, and the potential infringement of the rights holder’s legitimate interests. In doing so, the court refined and structured the relevant legal doctrines with greater precision. In this context, this article analyzes the interpretative standards and application framework of the German TDM provisions, focusing on the appellate judgment. It examines the relationship between TDM and generative AI training, the legal and technical requirements of the opt-out mechanism, and the balancing of interests under the three-step test. Furthermore, in light of the Korean copyright system, which largely relies on a general fair use clause, this article seeks to derive comparative insights from the German decision-particularly regarding purpose-oriented regulation, the practical realization of rights holders’ control, and a phased assessment structure that reflects technological realities-for the future design of domestic TDM norms.